• KIPS Transactions on Computer and Communication Systems
• /
• v.10 no.7
• /
• pp.191-198
• /
• 2021

Since the size of training dataset become large and the model is getting deeper to achieve high accuracy in deep learning, the deep neural network training requires a lot of computation and it takes too much time with a single node. Therefore, distributed deep learning is proposed to reduce the training time by distributing computation across multiple nodes. In this study, we propose hybrid allreduce strategy that considers the characteristics of each layer and communication and computational overlapping technique for synchronization of distributed deep learning. Since the convolution layer has fewer parameters than the fully-connected layer as well as it is located at the upper, only short overlapping time is allowed. Thus, butterfly allreduce is used to synchronize the convolution layer. On the other hand, fully-connecter layer is synchronized using ring all-reduce. The empirical experiment results on PyTorch with our proposed scheme shows that the proposed method reduced the training time by up to 33% compared to the baseline PyTorch.

• Journal of Korea Society of Industrial Information Systems
• /
• v.27 no.6
• /
• pp.41-49
• /
• 2022

The Internet Engineering Task Force (IETF) has standardized RPL (IPv6 Routing Protocol for Low-power Lossy Network) as a routing protocol for Low Power and Lossy Networks (LLNs), a low power loss network environment. RPL creates a route through an Objective Function (OF) suitable for the service required by LLNs and builds a Destination Oriented Directed Acyclic Graph (DODAG). Existing studies check the residual energy of each node and select a parent with the highest residual energy to build a DODAG, but the energy exhaustion of the parent can not avoid the network disconnection of the children nodes. Therefore, this paper proposes EC-RPL (Enhanced Connectivity-RPL), in which ta node leaves DODAG in advance when the remaining energy of the node falls below the specified energy threshold. The proposed protocol is implemented in Contiki, an open-source IoT operating system, and its performance is evaluated in Cooja simulator, and the number of control messages is compared using Foren6. Experimental results show that EC-RPL has 6.9% lower latency and 5.8% fewer control messages than the existing RPL, and the packet delivery rate is 1.7% higher.

• Journal of Internet Computing and Services
• /
• v.17 no.4
• /
• pp.11-17
• /
• 2016

In this paper, we proposed a multi-viewer system with multiple HD cameras based AVB(Audio Video Bridge) ethernet cable using IP networking, and FPGA(Xilinx Zynq 702) for bus safety systems. This AVB (IEEE802.1BA) system can be designed for the low latency based on FPGA, and transmit real-time with HD video and audio signals in a vehicle network. The proposed multi-viewer platform can multiplex H.264 video signals from 4 wide-angle HD cameras with existed ethernet 1Gbps. and 2-wire 100Mbps cables. The design of Zynq 702 based low latency to H.264 AVC CODEC was proposed for the minimization of time-delay in the HD video transmission of car area network, too. And the performance of PSNR(Peak Signal-to-noise-ratio) was analyzed with the reference model JM for encoding and decoding results in H.264 AVC CODEC. These PSNR values can be confirmed according the theoretical and HW result from the signal of H.264 AVC CODEC based on Zynq 702 the multi-viewer with multiple cameras. As a result, proposed AVB multi-viewer platform with multiple cameras can be used for the surveillance of audio and video around a bus for the safety due to the low latency of H.264 AVC CODEC design.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.52 no.1
• /
• pp.32-39
• /
• 2015

The voice communications with wireless backbone based networks are evolving into a packet switching VoIP systems. In those networks, a call processing scheme is required for management of subscribers and connection between them. A VoIP service scheme for those systems requires reliable subscriber management and connection establishment schemes, but the conventional call processing schemes based on the centralized server has lack of reliability. Thus, the mutual-backup architecture of SIP-servers is required to ensure efficient subscriber management and reliable VoIP call processing capability, and the synchronization and call processing schemes should be changed as the architecture is changed. In this paper, a mutual-backup architecture of SIP-servers is proposed for wireless backbone based networks. A message format for synchronization and information exchange between SIP servers is also proposed in the paper. This paper also proposes a FSM scheme for the fast call processing in unreliable networks to detect multiple servers at a time. The performance analysis results show that the mutual backup server architecture increases the call processing success rates than conventional centralized server architecture. Also, the FSM scheme provides the smaller call processing times than conventional SIP, and the time is not increased although the number of SIP servers in the networks is increased.

• Korean Journal of Construction Engineering and Management
• /
• v.6 no.4 s.26
• /
• pp.80-90
• /
• 2005

Activity durations retain probabilistic and stochastic natures due to diverse factors causing the delay or acceleration of activity completion. These natures make the final project duration to be a random variable. These factors are the major source of financial risk. Extending the Stochastic Project Scheduling Simulation system (SPSS) developed in previous research; this research presents a method to estimate how the final project duration behaves when activity durations change randomly. The final project cost is estimated by considering the fluctuation of indirect cost, which occurs due to the delay or acceleration of activity completion, along with direct cost assigned to an activity. The final project cost is estimated by considering how indirect cost behaves when activity duration change. The method quantifies the amount of contingency to cover the expected delay of project delivery. It is based on the quantitative analysis to obtain the descriptive statistics from the simulation outputs (final project durations). Existing deterministic scheduling method apply an arbitrary figures to the amount of delay contingency with uncertainty. However, the stochastic method developed in this research allows computing the amount of delay contingency with certainty and certain degree of confidence. An example project is used to illustrate the quantitative analysis method using simulation. When the statistical location and shape of probability distribution functions defining activity durations change, how the final project duration and cost behave are ascertained using automated sensitivity analysis method

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.41 no.1
• /
• pp.23-34
• /
• 2004

This paper presents architecture and control method of packet scheduler to guarantee QoS of high quality streaming services in high-speed packet-switched networks. Since streaming services need far more stringent QoS requirements than the typical sort of burst data applications, they should be guaranteed minimum bandwidth and end-to-end delay bound to each flow, regardless of the behavior of other flows. To meet these requirements, a packet scheduler isolate a flow from the undesirable effects of other flows and provides end-to-end delay guarantees for individual flow and divides stringently the available link bandwidth among flows sharing the link. Until now, many vendors are developing traffic management chips running at 10Gbps, but most of chips have drawbacks to support high quality streaming services. In this paper, we investigate the drawbacks of commercial TM chips and traffic characteristic of streaming services and present implementation frameworks of the proposed packet scheduler. Finally, we analyze the simulation results of the proposed scheduler.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.44 no.10
• /
• pp.119-126
• /
• 2007

Recently, the mobility of users and mobile communication technologies have developed rapidly. The users in this state also want to connect their devices and to receive services anywhere, anytime. Hierarchical Mobile IPv6 (HMIPv6) has been proposed by the Internet Engineering Task Force (IETF) to compensate for such problems as handover latency and signaling overhead when employing Mobile IPv6 (MIPv6). HMIPv6 supports micro-mobility within a domain and introduces a new entity, namely Mobility Anchor Point (MAP) as a local home agent. However, HMIPv6 has been found to cause longer handover latency when the inter-domain handover occurs. This is because a Mobile Node (MN) has to generate two addresses and register them to Home Agent (HA) a MAP, respectively. In order to solve such problems, we propose a scheme that an MN generates one address and registers it to HA for supporting fast handover during the inter-domain handover process. In the proposed scheme, the load of MAP and MAP domain is reduced because the number of MNs which are managed by MAP is decreased and the MAP does not perform proxy Neighbor Discovery Protocol (NDP) to intercept packets destined to MNs. We evaluate the performance of proposed scheme in comparison to HMIPv6 through the simulation and numerical analysis.

• Journal of Internet Computing and Services
• /
• v.3 no.1
• /
• pp.81-90
• /
• 2002

Large-scale distributed applications based on Internet and client/server applications have to deal with series of problems. Load balancing, unpredictable communication delays, and networking failures can be the example of the series of problems. Therefore. sophisticated applications such as teleconferencing, video-on-demand, and concurrent software engineering require an abstracted group communication, CORBA does not address these paradigms adequately. It mainly deals with point-to-point communication and does not support the development of reliable applications that include predictable behavior in distributed systems. In this paper, we present our design, implementation and performance analysis of the group communication using the CORBA-ORB. JAVA-RML and Socket based on distributed computing Performance analysis will be estimated latency-lime according to object increment, in case of group communication using ORB of CORBA the average is 14.5172msec, in case of group communication using RMI of Java the average is 21.4085msec, in case of group communication using socket the average is becoming 18.0714msec. Each group communication using multicast and UDP can be estimated 0.2735msec and 0.2157msec. The performance of the CORBA-ORB group communication is increased because of the increased object by the result of this research. This study can be applied to the fault-tolerant client/server system, group-ware. text retrieval system, and financial information systems.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.34 no.10B
• /
• pp.1050-1061
• /
• 2009

The Tactical Internet(TI) managed by Infantry Brigades is used for the purpose of sharing information of Command Control and Situation Awareness. When there are more than two destinations to transmit data in the TI system, a multicasting is utilized based on pre-defined multicast groups. However even in the case when a source node needs to send some messages like weathercast and attack alarm etc to only a part of Battalion or Brigades in a specific geographical region (destination region), the current TI multicasting protocol is designed to transmit the messages to the pre-defined group or all of the Battalion/Brigade nodes, resulting in inefficiency in terms of end-to-end delay and overhead. In this paper, we propose more efficient protocol for such cases, named as "Tactical Internet Geocasting (TIG)". The proposed scheme firstly checks whether the destination region belongs to one Battalion region or more than two Battalion regions using location information, and then performs a greedy forwarding from the source node to the destination region, followed by a local flooding inside of the destination region. With performance analysis and simulations using NS-2, TIG is compared to the current TI multicasting protocol (i.e., Simplified MDP) and the LBM (Location-based Multicast). The simulation results show that the proposed TIG is more efficient than both in terms of delay and network overhead.

• Journal of the Microelectronics and Packaging Society
• /
• v.16 no.1
• /
• pp.13-19
• /
• 2009

A 440MHz wireless and passive surface acoustic wave (SAW) based chemical sensor was developed on a $41^{\circ}YX\;LiNbO_3$ piezoelectric substrate for simultaneous measurement of $CO_2$ gas and relative humidity (RH) using a reflective delay line pattern as the sensor element. The reflective delay line is composed of an interdigital transducer (IDT) and several shorted grating reflectors. A Teflon AF 2400 and a hydrophilic $SiO_2$ layer were used as $CO_2$ and water vapor sensitive films. The coupling of mode (COM) modeling was conducted to determine optimal device parameters prior to fabrication. According to simulation results, the device was fabricated and then wirelessly measured using the network analyzer. The measured reflective coefficient $S_{11}$ in the time domain showed high signal/noise (S/N) ratio, small signal attenuation, and few spurious peaks. In the $CO_2$ and humidity testing, high sensitivity ($2^{\circ}/ppm$ for $CO_2$ detection and $7.45^{\circ}/%$RH for humidity sensing), good linearity and repeatability were observed in the $CO_2$ concentration ranges of $75{\sim}375ppm$ and humidity levels of $20{\sim}80%$RH. Temperature and humidity compensations were also investigated during the sensitivity evaluation process.